Securing identity for the age of AI agents with CrowdStrike

[00:00:00] The leading issue of agentic AI in businesses right now is ensuring agents act with compliance guidelines. And Denodo applies guardrails across your entire data estate. By aligning your company's data infrastructure under one system, these guardrails perform consistently across your platform. So start scaling your business and start with Denodo. Simply visit Denodo.com to learn more.

[00:00:33] How do you secure identity when the user requesting access could be an AI agent rather than an employee? Well, in this episode today, I'm going to be joined by the CTO for Europe at CrowdStrike. And he's going to explain how AI is changing both sides of the security equation.

[00:00:57] Attackers can research targets, write convincing phishing messages, analyze disclosed vulnerabilities, and abuse trusted software packages and do so at greater speed. Machine speed. So defenders can use the same technology to recognize patterns, automate responses, and give overstretched teams room to focus on some of that higher value work.

[00:01:21] But I also want to unlock some of the insights today from CrowdStrike's threat hunting report. Learn more about the growing number of non-human identities inside enterprises and why visibility must come before permissions and guardrails. So if your organization cannot say which models and agents are operating, what data they touch or what they cost, I'm hoping today's conversation will give you a useful place to begin.

[00:01:50] But enough from me. Let me introduce you to him now. So thank you for joining me on the podcast today. Can you tell everyone listening a little about who you are and what you do? Hey, Neil. Thanks so much for having me. Zeki Turedi. I'm the CLCTO for Europe at CrowdStrike. At CrowdStrike, I spend most of my time working with CISOs, CIOs, organizations, helping them build their cybersecurity strategy.

[00:02:17] And before I had the joys of joining CrowdStrike, I also worked as a digital forensics investigator and helping organizations build their own present intelligence and IRK abilities. Awesome. It's a pleasure to have you join me today. There's a lot of noise around AI at the moment, and especially AI as a tool is a phrase that seems to get bandied around a lot. Now, CrowdStrike, you describe AI as a tool, a target, and a force multiplier for modern adversaries.

[00:02:46] So can you give me a practical example of each, just so we don't lose anyone or leave anyone behind here today? Yeah, and I think it's a really good kind of example of where it becomes quite complex with AI. You know, there's so many places that AI really, really touches. And when we look into the adversaries, so first of all, you know, when we talk about a tool, it's the same kind of tool that everyone at home or in their own work environment that they're trying to utilize.

[00:03:15] Right. So you think most people are looking to utilize these AI tools to be able to have, you know, one up more efficiencies, be able to do those mundane tasks more rapidly. You know, many organizations are looking to bring these tools in to make sure they can scale, you know, be able to be more competitive. Funny enough, it's the exact same mindset when it comes to the threat actor. They know about these tools. They like to play with technology.

[00:03:44] You know, they are technologists. And guess what? They've been playing around with AI tools for quite some time. And, you know, really what we saw over the last 12 months is that the adversary that we know has become AI enabled. You know, they're using AI throughout their attack life cycles in ways to actually help them be more efficient.

[00:04:05] So as an example, you know, a long time ago, we would say it's really easy to identify a phishing email because there'll be spelling mistakes. There'll be grammatical errors and, you know, just won't look right. And a savvy person could easily determine it's fake versus real. Well, funnily enough, now with the use of AI, you know, these adversaries can send out huge amounts of really well-crafted emails targeting individuals.

[00:04:32] They can use AI to do, you know, background analysis and understand, you know, the company they're targeting and what kind of information might be more of interest to them and more likely to hit the link. So again, it's really enabling them. And then as a target, this is, again, just a similar theme we've seen through many years. You know, we as organizations, you know, are massively investing in technologies.

[00:04:59] And these technologies are just another entry point into our businesses. You know, a really good example of this is cloud. You know, over the last several years, organizations have been rapidly investing in cloud because it's good for their business. But this opens up another avenue for, you know, the malicious threat actors to target, to get access to that organization. And what we've seen is, again, as companies are rushing to invest in these new AI tools, they don't want to be left behind.

[00:05:28] But what they are forgetting about is security and they are leaving security behind. So these tools that they're implementing in their own organizations are now a soft and easy target to malicious actors. And then, of course, it is a force multiplier. You know, we've all used it. You know, we know that it can make, you know, complex tasks complete in a few moments with a little bit of tuning and tweaking. And we can do many tasks easily.

[00:05:57] And again, it is really scaling up that single adversary, blowing the bar to entry, really, should we say, to allow them to be able to be even more dangerous. Yes. And before you join me on the podcast, I was doing a little research and I came across your own research at CrowdStrike, which said that, well, it found that China-linked adversaries were exploiting vulnerabilities within 24 hours of public proof of concept code appearing, which blew my mind.

[00:06:24] But what does that shrinking response window mean for security teams? Yeah, again, it really means that organizations need to be a lot quicker and a lot more efficient operationally. You know, simply what it means is, you know, as an example for those not going to be familiar with the whole concept, this is when a vulnerability gets released into the public.

[00:06:47] So, you know, vendor finds out through various ways that they have vulnerability in their products, they will, you know, release a patch, a fix, and that fix goes into the wild. And, you know, everyone should be installing that patch, that fix immediately. But realistically, that's not what happens.

[00:07:11] You know, it's quite hard to do these patches and these updates on a continuous basis, especially with the sheer noise and amount of these patches we're seeing. And, you know, adversaries like the Chinese Nexus adversaries are aware that organizations are very, very slow to fixing problems in their environment.

[00:07:31] So rather than spending huge amounts of time trying to create a zero day, you know, try and find a bug, you've basically got a vendor telling you, hey, there's a problem here, but we fixed it. Reverse engineer the fix, and you very, very quickly start identifying the vulnerability, which you can then target en masse to organizations.

[00:07:52] And we have to remember that, you know, AI tools actually make it a lot easier for organizations, you know, for malicious organizations to do this as well. And elsewhere, you also found that North Korean linked actors were reportedly poisoning 131 trusted AI framework packages.

[00:08:12] The poisoning of AI models is something we don't talk about enough, but why are developer tools, models and software supply chains, why are they all becoming such attractive routes to organizations now? Again, there's many reasons for it. But if you think about it, if we stick on the AI topic, you know, everyone now believes themselves to be a fantastic software engineer.

[00:08:35] You know, you start utilizing an off-the-shelf AI model to do some development work, and it's going to start pulling down what it believes is to be the best and easiest packages. And a lot of these packages, these third-party packages are coming through package libraries and tools that unfortunately can be easily compromised.

[00:08:59] So we have a huge influx of people harnessing these third-party tools who may not actually understand where they're coming from and how to actually trust them or why they even utilize them in the first place. So first of all, we've seen that huge, huge boom. But on the other side, you know, we have to also remember that these third-party packages are the backbone of the internet. You know, we have huge amounts of infrastructure running off open source tools or third-party packages.

[00:09:29] And a lot of organizations haven't realized that they can be easily compromised and they haven't put in the process in place to understand the integrity of these packages. And are they directly linking them to the internet or are they keeping a local version that's safe, as an example? And, you know, the adversaries and the threat actors are aware of this. And it's not just North Korean actors. We're seeing a huge influx of criminal active groups as well.

[00:09:57] And the compromises may look quite small. Or, you know, let's just not talk about creating like back doors into an organization. In some cases, it's deploying crypto miners. It's deploying information stealers. But those information stealers can take extremely sensitive information and then be sold off to a neurocriminal group. So, again, they can be extremely damaging to an organization.

[00:10:25] And, of course, this year, all everyone's talking around is AI agents, agentic AI. And as businesses rush to deploy more and more AI agents, how does identity security change when the actor requesting might not even be a human employee? It's a massive talking point at the moment, isn't it?

[00:10:47] Yeah, and I think it's extremely important as well because more and more of the years go by, we're realizing how important identity and that identity infrastructure is the key to our organization. Everything relies on our identity infrastructure. And historically, it's kind of become stagnant or forgotten about, unfortunately, in most businesses.

[00:11:10] But if you have a strong identity layer and strong security and visibility into your identity layer, you have a perfect baseline to start bringing in new technologies like AI tools. The unfortunate scene that we see most of the time is that it's underinvested. You know, people aren't looking at identity from a cybersecurity perspective. And that means when you start adding things like AI agents, you've lost control. You have no visibility.

[00:11:39] You can't have any guardrails. You don't know who's accessing it, how they're accessing it. Is it human, non-human? Is it agentic identity? And these are all the very, very important questions we need to be able to quickly answer if we want to secure our enterprise networks. I completely agree with you. And I think this is only going to increase if we go looking forward to next year, for example. We're going to start seeing individuals having dozens of agents.

[00:12:07] And that number is only going to grow. And then with that, when you look at teams having the same and then the entire enterprise, it just grows rapidly. So what should an organization know about an agent before granting it access? And should its permissions differ from those of a person or a system that created it?

[00:12:26] I mean, going back on to what you were saying about the amount of agents per person, we're already seeing organizations that will have like 140, 150 agents per employee they have within their company. So it's not everywhere just yet, but that is, it's already happening. And we're seeing that noise and that sheer number of agents being used for organizations.

[00:12:48] And then the questions, you know, go back to kind of, you know, the questions organizations should be asking is, do they have enough visibility into what that agent's actually doing? Do they have visibility into what data that agent has access to? Do they actually understand how that agent was designed? Do they have a model? Do they have tracking of what's going in and out?

[00:13:10] Or, you know, so really it becomes more the most important fundamental when it comes to these new technologies is, do you have the visibility and transparency into what exactly they're doing throughout the whole life cycle? And if we have the visibility, we can then start putting in the guardrails we need, the policies within our organization and being able to remediate them.

[00:13:36] One of the kind of questions I ask to people quite often is, do you even know what models are being used in your organization? And, you know, the whole cost of the tokens being used in your organization. And a lot of time it's no, we have no visibility, but without that initial visibility, you can't put in the really important elements of policy guardrails and protection. And naturally we're talking a lot around the attackers and the sophisticated tools they've got at their disposal.

[00:14:06] And our news feeds will be full of horror stories of threats and attackers, but the defenders have very similar technology too, if not the same. So they're gaining a measurable advantage as well, rather than just processing more alerts for cybersecurity agents that are suffering with alert fatigue and looking at dashboards. Things are changing here for the good too, right? Yeah, exactly. And, you know, we're at, there's a fantastic opportunity right now, especially in cybersecurity.

[00:14:35] And I think a lot of people forget that cybersecurity has been investing in, you know, AI data science for a very, very long time. And we have now great datasets of all the different types of activities we're seeing over the last, you know, 15 years. And that data is so important to then be able to design the models of the future to be able to stop these types of incidents. But, you know, what we see is that organizations have the tools, the technologies available.

[00:15:06] You know, CrowdStrike, you know, harnesses, you know, the agentic capabilities to help organizations, you know, build an agentic stock. But what we need people to do is go, you know, it's time to move our humans from being reactive and actually change our model to be more proactive. And that's the journey that we're helping a lot of organizations do today. Like how can they re-team and remaster the humans they already have as part of their security teams to be, to work alongside the new agentic AI technologies.

[00:15:35] Of course, every threat report can generate urgency, but without changing behavior sometimes, because it's easy to get distracted and move back onto normal business. So what finding from this year's research, though, should cause leaders to maybe alter a process of control or investment? Anything that really stood out to you and did anything surprise you as well? I wouldn't say anything directly.

[00:16:01] I think it's the overall kind of story that I hope people got or will get if they read the report. You know, we have to remember that, you know, on one side, we still have the human threat actor. The human threat actor is more advanced than ever before. They're more sophisticated. They're quicker. They're smarter. And then, of course, we have on the other element, the AI confusion. You know, AI brings a lot of different changes into our environment today.

[00:16:29] Really, what my recommendation for is a lot of organizations is to really get the security core rectified. You know, the investments that most organizations have been reluctant to do. You know, we spoke about securing identity as a great example. You know, if we get the core in a good place when it comes to securing our enterprise organizations, then things like AI is not a problem for us.

[00:16:57] You know, as an example, you know, if we are able to remediate vulnerabilities and patches in our organization in a timely manner, we shouldn't be scared about, you know, frontier AI or Chinese nation states being able to identify vulnerabilities rapidly. You know, if we've got our team set up to utilize the technologies for the eugentic stock, we can deal with the speed of efficiencies of AI tools or, you know, human-enabled, AI-enabled adversaries.

[00:17:27] I want to put you on the spot now, then. If you were testing one part of an organization against AI-supported adversaries, would you start with identity? Because that's what everyone's talking about at the moment. Or vulnerability response, developer tooling, or the software supply chain? Where would you start and why? Unfortunately, just to kind of really, really confuse stuff, we have to focus on all of them right now.

[00:17:54] There's no, there's no, they're all areas of, you know, of keen interest to the adversary and also large weaknesses into organizations. And the reality is, you know, the instance that we see would actually touch upon all of them today. You know, so my biggest recommendation is probably don't leave it until next week. Get started, work with professionals and partners and test yourself as much as possible.

[00:18:20] And outside of that, there's a lot of myths and misconceptions that maybe frustrate you when you're scrolling down your LinkedIn newsfeed. Some untruths, some stretches of the truth, etc. I'm curious if we could lay any myths to rest today. Is that any big frustrations that you would like to quietly dispel once and for all? I think we need to not forget the human behind the scenes. You know, we are dealing with human actors.

[00:18:48] These are, you know, coming from various different organizations. And, you know, the reality is whatever tool they're utilizing, we can stop them. You know, we do every single day. And I think we sometimes are so focused on the organizations or the incidents that were, you know, targeting an organization or made, you know, an organization a victim. We forget about, you know, the hundreds of thousands successful, you know, preventions that happen every single day.

[00:19:19] So it looks daunting. It can sound scary. We are definitely at this huge technology boom. But the reality is, you know, organizations are keeping themselves secure and safe throughout the globe as long as they are prepared and have the right tools and partners for it. And the industry has greatly evolved over the last few years. So what makes you optimistic about the future? Maybe even excites you about the work that you're doing at CrowdStrike and what you're seeing and hearing there.

[00:19:48] Anything in particular stand out there? Yeah, I think, again, it's AI, AI, AI. Yeah. You know, we have to remember, again, we talk a lot about the doom and gloom and about AI and, you know, the chaos it might cause from a cybersecurity perspective. But as you rightly mentioned, these are the exact same tools that defenders now have access to. And we have so much more data. Like we've been learning about threat actors, we've been tracking threat actors for a considerable amount of time.

[00:20:17] And that data with the harnessing of the technologies we have available to us today means that defenders are going to be way more capable than ever before. And that is just growing. So, yeah, I'm excited about this industry. I'm excited about, you know, artificial intelligence and generative AI and what it brings to us. And I think we're in a great place at the moment. Love that. And I'm hoping your passion and excitement will rub off on people listening.

[00:20:44] I'm going to send a link on the show notes for CrowdStrike's 2026 threat hunting report. Please, this is your call to action. Go out there. Have a read through and make some changes. But for anyone wanting to connect with you, your team or find out more information about all things CrowdStrike, anywhere else you'd point everyone listening? Just head to CrowdStrike.com. We've got loads of information.

[00:21:08] So if you're an amateur looking to understand more about cyber, we've got plenty of reports all the way up to if you're an advanced user, we've got some very, very technical reports for our blog as well. Awesome. Well, I will add links to everything, including the blog and indeed your own LinkedIn profile as well. Love to keep this conversation going. Hopefully we can see some positive changes out there too as a result of this conversation.

[00:21:35] But more than anything, thank you for taking the time out of a very busy day to come and speak with me today. I appreciate your time. Thanks, Neil. Thanks, everyone. So many big messages from today's episode, such as AI doesn't make familiar security disciplines obsolete. But it does make weak identity control, slow patching and poor supply chain visibility much harder to ignore.

[00:22:01] So before adding hundreds of agents per employee, an organisation needs to know what each of those agents can access, what model supports it, what information moves in and out, and what can stop it when behavior changes and drift occurs? Well, the hopeful part is that defenders are gaining speed too. Years of threat data can support faster detection and response.

[00:22:31] And security professionals are spending less time processing repetitive alerts, suffering from alert fatigue, staring at dashboards. That's got to be a good thing. So a big thank you to my guests for bringing some welcome balance to a conversation often dominated by fear. So you can find further research and technical guidance at crowdstrike.com. But over to you, are your AI agents entering a well-managed identity system?

[00:22:59] Or are they exposing problems that were already waiting to be found? As always, techtalksnetwork.com. Nice and easy to get hold of. I'd love to hear your thoughts on anything we've talked about today. But it's quitting time for me. Time for me to go now. I'll be back again real soon with another guest. Speak with you then. Bye for now.